On October 21, Solana retired the consensus it launched with and kept producing blocks through the swap. The 150 ms finality headline is Anza's target; the day-one readings outlets pulled from the Foundation's dashboard were 202 milliseconds and 6.6 seconds.
At about 13:36 UTC on October 21, epoch 1064 began on Solana mainnet and a countdown started: 5,000 more slots under TowerBFT, the consensus the chain had run since March 2020. Then slot 459,652,999 arrived. Votes from 446 of the network's 654 validators certified it as the Alpenglow genesis block; at slot 459,653,000 consensus belonged to the new protocol. The Solana Foundation puts the handoff at about 900 milliseconds.
No halt, no restart, and Unchained found no sign of a rollback.
Three days on, the question is not whether the swap worked. It did. The question is what the new number means.
One Slot, No Rollback
Anza (@anza_xyz), which builds the Agave validator client, says it took Alpenglow from a May 2025 white paper to mainnet in under 18 months.
The migration was the scary part. Two days before the switch, a Pumpkin's Pool readiness tracker, relayed by SolanaFloor (@SolanaFloor), flagged 45% of stake as not ready. By the morning of October 20 that was down to 151 validators holding 20.3%. The genesis certificate needed 82%.
There was one road through: three Agave-based clients at v4.3.2. Firedancer and Frankendancer were not supported for the migration; Galaxy's engineers, per Solana Compass, told Firedancer operators to fail over to Agave and return afterwards, since Firedancer 26.10 supports Alpenglow.
Solana did not stop to upgrade; it voted a new consensus into existence while the old one was still producing blocks. Triton One (@triton_one) warned integrators: "There's no going back to the old consensus afterwards."
https://x.com/anza_xyz/status/2112905478921736411
Then the gauges started talking.
Which Finality Got Faster
Anza's post said finality was "reduced from 12.8 seconds to 150 milliseconds." Jacob Creech (@jacobvcreech), the Solana Foundation's technology VP, compressed it: "Vote transactions -> 0, Finality -> 150ms, TowerBFT -> Gone."
https://x.com/jacobvcreech/status/2112905746736369967
Cointelegraph, citing the Foundation's dashboard "at press time," reported 6.6 seconds of finality and 4,721 TPS. AltcoinBuzz, reading the same dashboard on October 21 and 22, reported a latest observed finality of 202 ms and called it a single reading, not an average. I tried the dashboard on October 24 and got a 404.
My read: both can be true. TowerBFT finalized 32 slots after a block, and Solana has run 200 ms slots since October 9, so old finality just before the switch was about 6.4 seconds (my math: 32 × 200 ms). A reading taken minutes after the handoff would still be mostly TowerBFT.
The sharper question is which finality anyone was waiting for. Under TowerBFT, "confirmed" (a two-thirds stake vote) landed in roughly 500–600 ms; "finalized" took the full 12.8 seconds. Alpenglow collapses the two into one state, targeted at 150 ms.
The 85x headline compares the slowest old setting with the fastest new one (my math: 12.8 s ÷ 0.15 s).
Worked example: an app acting on "confirmed" went from about 550 ms to the 150–200 ms Anza's Max Resnick told Decrypt users would feel, roughly 3x. One waiting for "finalized" went from 12.8 seconds to the same 150–200 ms. Decrypt names exchanges, banks and merchants as the beneficiaries; those are the parties that wait for finalized.
The Thinner Shield
The speed is bought with a different fault model. The steelman is that it may be a better one.
- Malicious stake: TowerBFT tolerated up to a third of stake misbehaving. Alpenglow's safety holds to 20% adversarial stake, with liveness surviving another 20% offline. Galaxy, per Solana Compass, calls it a cut from 33% to 20%; Anza calls it surviving 40% failing at once.
- Slashing: double-voting penalties are, in Galaxy's words, "well defined but deliberately not enabled in the first iteration."
- Half an upgrade: Rotor, the relay layer meant to replace Turbine, is "not yet scheduled."
- The target is a target: the Foundation calls 150 ms an expectation measured against a past stake distribution, not a constant.
- Bigger blocks next: Jump's Firedancer team wants to remove the 60 million compute-unit cap now that slow validators can skip-vote on blocks they can't process. Anza's Roger Wattenhofer likes the idea and warned it could centralize block production.
Alpenglow bought its speed with a smaller margin against malicious stake and a bigger one against absent stake. In Anza's favor: nine migrations at scale and 115 security researchers before mainnet, per its status page, and Helius notes no optimistically confirmed block has ever been rolled back.
The Bottom Line
The market shrugged. SOL was $103.51 as of October 23, 64.7% below its January 2025 high of $293.31, per CoinGecko. US spot Solana ETFs lost $24.8 million in the week of October 5–9 per SoSoValue data; by Coinboss's daily table, 14 of October's 16 sessions through October 22 were outflows.
My position: the engineering is the real story, executed about as cleanly as a live consensus swap can be. The 150 ms figure is still a claim. A week of dashboard readings with a distribution would make it a fact; one 202 ms print does not. I'd watch that, the slashing switch and a Rotor date before the price.
Solana changed engines without landing. Now the gauges have to agree.
